1 // SPDX-License-Identifier: GPL-2.0-only 2 /* Atlantic Network Driver 3 * 4 * Copyright (C) 2014-2019 aQuantia Corporation 5 * Copyright (C) 2019-2020 Marvell International Ltd. 6 */ 7 8 /* File aq_ring.c: Definition of functions for Rx/Tx rings. */ 9 10 #include "aq_ring.h" 11 #include "aq_nic.h" 12 #include "aq_hw.h" 13 #include "aq_hw_utils.h" 14 #include "aq_ptp.h" 15 16 #include <linux/netdevice.h> 17 #include <linux/etherdevice.h> 18 19 static inline void aq_free_rxpage(struct aq_rxpage *rxpage, struct device *dev) 20 { 21 unsigned int len = PAGE_SIZE << rxpage->order; 22 23 dma_unmap_page(dev, rxpage->daddr, len, DMA_FROM_DEVICE); 24 25 /* Drop the ref for being in the ring. */ 26 __free_pages(rxpage->page, rxpage->order); 27 rxpage->page = NULL; 28 } 29 30 static int aq_get_rxpage(struct aq_rxpage *rxpage, unsigned int order, 31 struct device *dev) 32 { 33 struct page *page; 34 int ret = -ENOMEM; 35 dma_addr_t daddr; 36 37 page = dev_alloc_pages(order); 38 if (unlikely(!page)) 39 goto err_exit; 40 41 daddr = dma_map_page(dev, page, 0, PAGE_SIZE << order, 42 DMA_FROM_DEVICE); 43 44 if (unlikely(dma_mapping_error(dev, daddr))) 45 goto free_page; 46 47 rxpage->page = page; 48 rxpage->daddr = daddr; 49 rxpage->order = order; 50 rxpage->pg_off = 0; 51 52 return 0; 53 54 free_page: 55 __free_pages(page, order); 56 57 err_exit: 58 return ret; 59 } 60 61 static int aq_get_rxpages(struct aq_ring_s *self, struct aq_ring_buff_s *rxbuf, 62 int order) 63 { 64 int ret; 65 66 if (rxbuf->rxdata.page) { 67 /* One means ring is the only user and can reuse */ 68 if (page_ref_count(rxbuf->rxdata.page) > 1) { 69 /* Try reuse buffer */ 70 rxbuf->rxdata.pg_off += AQ_CFG_RX_FRAME_MAX; 71 if (rxbuf->rxdata.pg_off + AQ_CFG_RX_FRAME_MAX <= 72 (PAGE_SIZE << order)) { 73 self->stats.rx.pg_flips++; 74 } else { 75 /* Buffer exhausted. We have other users and 76 * should release this page and realloc 77 */ 78 aq_free_rxpage(&rxbuf->rxdata, 79 aq_nic_get_dev(self->aq_nic)); 80 self->stats.rx.pg_losts++; 81 } 82 } else { 83 rxbuf->rxdata.pg_off = 0; 84 self->stats.rx.pg_reuses++; 85 } 86 } 87 88 if (!rxbuf->rxdata.page) { 89 ret = aq_get_rxpage(&rxbuf->rxdata, order, 90 aq_nic_get_dev(self->aq_nic)); 91 return ret; 92 } 93 94 return 0; 95 } 96 97 static struct aq_ring_s *aq_ring_alloc(struct aq_ring_s *self, 98 struct aq_nic_s *aq_nic) 99 { 100 int err = 0; 101 102 self->buff_ring = 103 kcalloc(self->size, sizeof(struct aq_ring_buff_s), GFP_KERNEL); 104 105 if (!self->buff_ring) { 106 err = -ENOMEM; 107 goto err_exit; 108 } 109 self->dx_ring = dma_alloc_coherent(aq_nic_get_dev(aq_nic), 110 self->size * self->dx_size, 111 &self->dx_ring_pa, GFP_KERNEL); 112 if (!self->dx_ring) { 113 err = -ENOMEM; 114 goto err_exit; 115 } 116 117 err_exit: 118 if (err < 0) { 119 aq_ring_free(self); 120 self = NULL; 121 } 122 123 return self; 124 } 125 126 struct aq_ring_s *aq_ring_tx_alloc(struct aq_ring_s *self, 127 struct aq_nic_s *aq_nic, 128 unsigned int idx, 129 struct aq_nic_cfg_s *aq_nic_cfg) 130 { 131 int err = 0; 132 133 self->aq_nic = aq_nic; 134 self->idx = idx; 135 self->size = aq_nic_cfg->txds; 136 self->dx_size = aq_nic_cfg->aq_hw_caps->txd_size; 137 138 self = aq_ring_alloc(self, aq_nic); 139 if (!self) { 140 err = -ENOMEM; 141 goto err_exit; 142 } 143 144 err_exit: 145 if (err < 0) { 146 aq_ring_free(self); 147 self = NULL; 148 } 149 150 return self; 151 } 152 153 struct aq_ring_s *aq_ring_rx_alloc(struct aq_ring_s *self, 154 struct aq_nic_s *aq_nic, 155 unsigned int idx, 156 struct aq_nic_cfg_s *aq_nic_cfg) 157 { 158 int err = 0; 159 160 self->aq_nic = aq_nic; 161 self->idx = idx; 162 self->size = aq_nic_cfg->rxds; 163 self->dx_size = aq_nic_cfg->aq_hw_caps->rxd_size; 164 self->page_order = fls(AQ_CFG_RX_FRAME_MAX / PAGE_SIZE + 165 (AQ_CFG_RX_FRAME_MAX % PAGE_SIZE ? 1 : 0)) - 1; 166 167 if (aq_nic_cfg->rxpageorder > self->page_order) 168 self->page_order = aq_nic_cfg->rxpageorder; 169 170 self = aq_ring_alloc(self, aq_nic); 171 if (!self) { 172 err = -ENOMEM; 173 goto err_exit; 174 } 175 176 err_exit: 177 if (err < 0) { 178 aq_ring_free(self); 179 self = NULL; 180 } 181 182 return self; 183 } 184 185 struct aq_ring_s * 186 aq_ring_hwts_rx_alloc(struct aq_ring_s *self, struct aq_nic_s *aq_nic, 187 unsigned int idx, unsigned int size, unsigned int dx_size) 188 { 189 struct device *dev = aq_nic_get_dev(aq_nic); 190 size_t sz = size * dx_size + AQ_CFG_RXDS_DEF; 191 192 memset(self, 0, sizeof(*self)); 193 194 self->aq_nic = aq_nic; 195 self->idx = idx; 196 self->size = size; 197 self->dx_size = dx_size; 198 199 self->dx_ring = dma_alloc_coherent(dev, sz, &self->dx_ring_pa, 200 GFP_KERNEL); 201 if (!self->dx_ring) { 202 aq_ring_free(self); 203 return NULL; 204 } 205 206 return self; 207 } 208 209 int aq_ring_init(struct aq_ring_s *self) 210 { 211 self->hw_head = 0; 212 self->sw_head = 0; 213 self->sw_tail = 0; 214 215 return 0; 216 } 217 218 static inline bool aq_ring_dx_in_range(unsigned int h, unsigned int i, 219 unsigned int t) 220 { 221 return (h < t) ? ((h < i) && (i < t)) : ((h < i) || (i < t)); 222 } 223 224 void aq_ring_update_queue_state(struct aq_ring_s *ring) 225 { 226 if (aq_ring_avail_dx(ring) <= AQ_CFG_SKB_FRAGS_MAX) 227 aq_ring_queue_stop(ring); 228 else if (aq_ring_avail_dx(ring) > AQ_CFG_RESTART_DESC_THRES) 229 aq_ring_queue_wake(ring); 230 } 231 232 void aq_ring_queue_wake(struct aq_ring_s *ring) 233 { 234 struct net_device *ndev = aq_nic_get_ndev(ring->aq_nic); 235 236 if (__netif_subqueue_stopped(ndev, 237 AQ_NIC_RING2QMAP(ring->aq_nic, 238 ring->idx))) { 239 netif_wake_subqueue(ndev, 240 AQ_NIC_RING2QMAP(ring->aq_nic, ring->idx)); 241 ring->stats.tx.queue_restarts++; 242 } 243 } 244 245 void aq_ring_queue_stop(struct aq_ring_s *ring) 246 { 247 struct net_device *ndev = aq_nic_get_ndev(ring->aq_nic); 248 249 if (!__netif_subqueue_stopped(ndev, 250 AQ_NIC_RING2QMAP(ring->aq_nic, 251 ring->idx))) 252 netif_stop_subqueue(ndev, 253 AQ_NIC_RING2QMAP(ring->aq_nic, ring->idx)); 254 } 255 256 bool aq_ring_tx_clean(struct aq_ring_s *self) 257 { 258 struct device *dev = aq_nic_get_dev(self->aq_nic); 259 unsigned int budget; 260 261 for (budget = AQ_CFG_TX_CLEAN_BUDGET; 262 budget && self->sw_head != self->hw_head; budget--) { 263 struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head]; 264 265 if (likely(buff->is_mapped)) { 266 if (unlikely(buff->is_sop)) { 267 if (!buff->is_eop && 268 buff->eop_index != 0xffffU && 269 (!aq_ring_dx_in_range(self->sw_head, 270 buff->eop_index, 271 self->hw_head))) 272 break; 273 274 dma_unmap_single(dev, buff->pa, buff->len, 275 DMA_TO_DEVICE); 276 } else { 277 dma_unmap_page(dev, buff->pa, buff->len, 278 DMA_TO_DEVICE); 279 } 280 } 281 282 if (unlikely(buff->is_eop)) { 283 ++self->stats.tx.packets; 284 self->stats.tx.bytes += buff->skb->len; 285 286 dev_kfree_skb_any(buff->skb); 287 } 288 buff->pa = 0U; 289 buff->eop_index = 0xffffU; 290 self->sw_head = aq_ring_next_dx(self, self->sw_head); 291 } 292 293 return !!budget; 294 } 295 296 static void aq_rx_checksum(struct aq_ring_s *self, 297 struct aq_ring_buff_s *buff, 298 struct sk_buff *skb) 299 { 300 if (!(self->aq_nic->ndev->features & NETIF_F_RXCSUM)) 301 return; 302 303 if (unlikely(buff->is_cso_err)) { 304 ++self->stats.rx.errors; 305 skb->ip_summed = CHECKSUM_NONE; 306 return; 307 } 308 if (buff->is_ip_cso) { 309 __skb_incr_checksum_unnecessary(skb); 310 } else { 311 skb->ip_summed = CHECKSUM_NONE; 312 } 313 314 if (buff->is_udp_cso || buff->is_tcp_cso) 315 __skb_incr_checksum_unnecessary(skb); 316 } 317 318 #define AQ_SKB_ALIGN SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) 319 int aq_ring_rx_clean(struct aq_ring_s *self, 320 struct napi_struct *napi, 321 int *work_done, 322 int budget) 323 { 324 struct net_device *ndev = aq_nic_get_ndev(self->aq_nic); 325 bool is_rsc_completed = true; 326 int err = 0; 327 328 for (; (self->sw_head != self->hw_head) && budget; 329 self->sw_head = aq_ring_next_dx(self, self->sw_head), 330 --budget, ++(*work_done)) { 331 struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head]; 332 bool is_ptp_ring = aq_ptp_ring(self->aq_nic, self); 333 struct aq_ring_buff_s *buff_ = NULL; 334 struct sk_buff *skb = NULL; 335 unsigned int next_ = 0U; 336 unsigned int i = 0U; 337 u16 hdr_len; 338 339 if (buff->is_cleaned) 340 continue; 341 342 if (!buff->is_eop) { 343 buff_ = buff; 344 do { 345 next_ = buff_->next, 346 buff_ = &self->buff_ring[next_]; 347 is_rsc_completed = 348 aq_ring_dx_in_range(self->sw_head, 349 next_, 350 self->hw_head); 351 352 if (unlikely(!is_rsc_completed)) 353 break; 354 355 buff->is_error |= buff_->is_error; 356 buff->is_cso_err |= buff_->is_cso_err; 357 358 } while (!buff_->is_eop); 359 360 if (!is_rsc_completed) { 361 err = 0; 362 goto err_exit; 363 } 364 if (buff->is_error || 365 (buff->is_lro && buff->is_cso_err)) { 366 buff_ = buff; 367 do { 368 next_ = buff_->next, 369 buff_ = &self->buff_ring[next_]; 370 371 buff_->is_cleaned = true; 372 } while (!buff_->is_eop); 373 374 ++self->stats.rx.errors; 375 continue; 376 } 377 } 378 379 if (buff->is_error) { 380 ++self->stats.rx.errors; 381 continue; 382 } 383 384 dma_sync_single_range_for_cpu(aq_nic_get_dev(self->aq_nic), 385 buff->rxdata.daddr, 386 buff->rxdata.pg_off, 387 buff->len, DMA_FROM_DEVICE); 388 389 /* for single fragment packets use build_skb() */ 390 if (buff->is_eop && 391 buff->len <= AQ_CFG_RX_FRAME_MAX - AQ_SKB_ALIGN) { 392 skb = build_skb(aq_buf_vaddr(&buff->rxdata), 393 AQ_CFG_RX_FRAME_MAX); 394 if (unlikely(!skb)) { 395 err = -ENOMEM; 396 goto err_exit; 397 } 398 if (is_ptp_ring) 399 buff->len -= 400 aq_ptp_extract_ts(self->aq_nic, skb, 401 aq_buf_vaddr(&buff->rxdata), 402 buff->len); 403 skb_put(skb, buff->len); 404 page_ref_inc(buff->rxdata.page); 405 } else { 406 skb = napi_alloc_skb(napi, AQ_CFG_RX_HDR_SIZE); 407 if (unlikely(!skb)) { 408 err = -ENOMEM; 409 goto err_exit; 410 } 411 if (is_ptp_ring) 412 buff->len -= 413 aq_ptp_extract_ts(self->aq_nic, skb, 414 aq_buf_vaddr(&buff->rxdata), 415 buff->len); 416 417 hdr_len = buff->len; 418 if (hdr_len > AQ_CFG_RX_HDR_SIZE) 419 hdr_len = eth_get_headlen(skb->dev, 420 aq_buf_vaddr(&buff->rxdata), 421 AQ_CFG_RX_HDR_SIZE); 422 423 memcpy(__skb_put(skb, hdr_len), aq_buf_vaddr(&buff->rxdata), 424 ALIGN(hdr_len, sizeof(long))); 425 426 if (buff->len - hdr_len > 0) { 427 skb_add_rx_frag(skb, 0, buff->rxdata.page, 428 buff->rxdata.pg_off + hdr_len, 429 buff->len - hdr_len, 430 AQ_CFG_RX_FRAME_MAX); 431 page_ref_inc(buff->rxdata.page); 432 } 433 434 if (!buff->is_eop) { 435 buff_ = buff; 436 i = 1U; 437 do { 438 next_ = buff_->next, 439 buff_ = &self->buff_ring[next_]; 440 441 dma_sync_single_range_for_cpu( 442 aq_nic_get_dev(self->aq_nic), 443 buff_->rxdata.daddr, 444 buff_->rxdata.pg_off, 445 buff_->len, 446 DMA_FROM_DEVICE); 447 skb_add_rx_frag(skb, i++, 448 buff_->rxdata.page, 449 buff_->rxdata.pg_off, 450 buff_->len, 451 AQ_CFG_RX_FRAME_MAX); 452 page_ref_inc(buff_->rxdata.page); 453 buff_->is_cleaned = 1; 454 455 buff->is_ip_cso &= buff_->is_ip_cso; 456 buff->is_udp_cso &= buff_->is_udp_cso; 457 buff->is_tcp_cso &= buff_->is_tcp_cso; 458 buff->is_cso_err |= buff_->is_cso_err; 459 460 } while (!buff_->is_eop); 461 } 462 } 463 464 if (buff->is_vlan) 465 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 466 buff->vlan_rx_tag); 467 468 skb->protocol = eth_type_trans(skb, ndev); 469 470 aq_rx_checksum(self, buff, skb); 471 472 skb_set_hash(skb, buff->rss_hash, 473 buff->is_hash_l4 ? PKT_HASH_TYPE_L4 : 474 PKT_HASH_TYPE_NONE); 475 /* Send all PTP traffic to 0 queue */ 476 skb_record_rx_queue(skb, 477 is_ptp_ring ? 0 478 : AQ_NIC_RING2QMAP(self->aq_nic, 479 self->idx)); 480 481 ++self->stats.rx.packets; 482 self->stats.rx.bytes += skb->len; 483 484 napi_gro_receive(napi, skb); 485 } 486 487 err_exit: 488 return err; 489 } 490 491 void aq_ring_hwts_rx_clean(struct aq_ring_s *self, struct aq_nic_s *aq_nic) 492 { 493 #if IS_REACHABLE(CONFIG_PTP_1588_CLOCK) 494 while (self->sw_head != self->hw_head) { 495 u64 ns; 496 497 aq_nic->aq_hw_ops->extract_hwts(aq_nic->aq_hw, 498 self->dx_ring + 499 (self->sw_head * self->dx_size), 500 self->dx_size, &ns); 501 aq_ptp_tx_hwtstamp(aq_nic, ns); 502 503 self->sw_head = aq_ring_next_dx(self, self->sw_head); 504 } 505 #endif 506 } 507 508 int aq_ring_rx_fill(struct aq_ring_s *self) 509 { 510 unsigned int page_order = self->page_order; 511 struct aq_ring_buff_s *buff = NULL; 512 int err = 0; 513 int i = 0; 514 515 if (aq_ring_avail_dx(self) < min_t(unsigned int, AQ_CFG_RX_REFILL_THRES, 516 self->size / 2)) 517 return err; 518 519 for (i = aq_ring_avail_dx(self); i--; 520 self->sw_tail = aq_ring_next_dx(self, self->sw_tail)) { 521 buff = &self->buff_ring[self->sw_tail]; 522 523 buff->flags = 0U; 524 buff->len = AQ_CFG_RX_FRAME_MAX; 525 526 err = aq_get_rxpages(self, buff, page_order); 527 if (err) 528 goto err_exit; 529 530 buff->pa = aq_buf_daddr(&buff->rxdata); 531 buff = NULL; 532 } 533 534 err_exit: 535 return err; 536 } 537 538 void aq_ring_rx_deinit(struct aq_ring_s *self) 539 { 540 if (!self) 541 goto err_exit; 542 543 for (; self->sw_head != self->sw_tail; 544 self->sw_head = aq_ring_next_dx(self, self->sw_head)) { 545 struct aq_ring_buff_s *buff = &self->buff_ring[self->sw_head]; 546 547 aq_free_rxpage(&buff->rxdata, aq_nic_get_dev(self->aq_nic)); 548 } 549 550 err_exit:; 551 } 552 553 void aq_ring_free(struct aq_ring_s *self) 554 { 555 if (!self) 556 goto err_exit; 557 558 kfree(self->buff_ring); 559 560 if (self->dx_ring) 561 dma_free_coherent(aq_nic_get_dev(self->aq_nic), 562 self->size * self->dx_size, self->dx_ring, 563 self->dx_ring_pa); 564 565 err_exit:; 566 } 567